U.S. patent number 8,401,224 [Application Number 12/774,512] was granted by the patent office on 2013-03-19 for hidden image signalling.
This patent grant is currently assigned to Digimarc Corporation. The grantee listed for this patent is Geoffrey B. Rhoads. Invention is credited to Geoffrey B. Rhoads.
United States Patent |
8,401,224 |
Rhoads |
March 19, 2013 |
Hidden image signalling
Abstract
An image is encoded to define one or more spatial regions that
can be sensed by a suitably-equipped mobile device (e.g., a smart
phone), but are imperceptible to humans. When such a mobile device
senses one of these regions, it takes an action in response (e.g.,
rendering an associated tone, playing linked video, etc.). The
regions may overlap in layered fashion. One form of encoding
employs modification of the color content of the image at higher
spatial frequencies, where human vision is not acute. In a
particular embodiment, the encoding comprises altering a transform
domain representation of the image by adding signal energy in a
first chrominance channel, where the added signal energy falls
primarily within a segmented arc region in a transform domain
space.
Inventors: |
Rhoads; Geoffrey B. (West Linn,
OR) |
Applicant: |
Name |
City |
State |
Country |
Type |
Rhoads; Geoffrey B. |
West Linn |
OR |
US |
|
|
Assignee: |
Digimarc Corporation
(Beaverton, OR)
|
Family
ID: |
44901956 |
Appl.
No.: |
12/774,512 |
Filed: |
May 5, 2010 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20110274310 A1 |
Nov 10, 2011 |
|
Current U.S.
Class: |
382/100; 382/276;
382/217; 382/162 |
Current CPC
Class: |
H04N
1/32309 (20130101); H04N 1/32 (20130101); G06T
1/00 (20130101); H04N 1/3216 (20130101); G06T
1/005 (20130101); H04N 1/32187 (20130101); H04N
2201/3264 (20130101); H04N 2201/3249 (20130101); G06T
2201/0052 (20130101); G06T 2201/0601 (20130101) |
Current International
Class: |
G06K
9/00 (20060101) |
Field of
Search: |
;382/100,162,165,217,276,280 |
References Cited
[Referenced By]
U.S. Patent Documents
Other References
Tsui et al. "Color Image Watermarking Using Multidimensional
Fourier Transforms." IEEE Transactions on Information Forensics and
Security, vol. 3, No. 1, Mar. 2008, pp. 16-28. cited by examiner
.
Surachat et al. "Pixel-wise based Digital Watermarking Using Weiner
Filter in Chrominance Channel." 9th International Symposium on
Communications and Information Technology, Sep. 28, 2009, pp.
887-892. cited by examiner .
Liu et al. "Robust and Transparent Watermarking Scheme for Colour
Images." Image Processing, IET, vol. 3, No. 4, Aug. 2009, pp.
228-242. cited by examiner .
Seo et al. "Color Images Watermarking of Multi-Level Structure for
Multimedia Services." International Conference on Convergence
Information Technology, Nov. 21, 2007, pp. 854-860. cited by
examiner .
Bors et al. "Image Watermarking Using DCT Domain Constraints."
Proceedings, International Conference on Image Processing, vol. 3,
Sep. 16, 1996, pp. 231-234. cited by examiner .
Solachidis et al. "Circularly Symmetric Watermark Embedding in 2-D
DFT Domain." IEEE Transactions on Image Processing, vol. 10, No.
11, Nov. 2001, pp. 1741-1753. cited by examiner .
Funk. "Image Watermarking in the Fourier Domain Based on Global
Features of Concentric Ring Areas." Proceedings of SPIE 4675,
Security and Watermarking of Multimedia Contents IV, 596, Apr. 29,
2002, pp. 596-599. cited by examiner .
International Search Report and Written Opinion dated Sep. 9, 2011,
in PCT/US11/34829. cited by applicant.
|
Primary Examiner: Chang; Jon
Attorney, Agent or Firm: Digimarc Corporation
Claims
I claim:
1. A method of marking color imagery employing a device having a
processor, configured by software stored in a memory, to perform at
least certain acts of the method, the method comprising the acts:
within imagery in a spatial or pixel domain, identifying at least
one two-dimensional image sub-region; the device processor encoding
one or more chroma keys in said sub-region by altering a transform
domain representation thereof, said altering comprising adding
image signal energy in a first chrominance channel, said added
image signal energy falling primarily within a segmented arc region
in a transform domain space; and transmitting, storing or
presenting the altered imagery, together with different auxiliary
information; wherein the encoded chroma key(s), in conjunction with
the auxiliary information, serve to cooperatively define a response
to said image sub-region when sensed by a mobile device including
an image sensor.
2. The method of claim 1 in which the segmented arc region has a
lower spatial frequency of at least 10 cycles per degree.
3. The method of claim 1 in which the segmented arc region has a
lower spatial frequency of at least 20 cycles per degree.
4. The method of claim 1 in which the segmented arc region has a
lower spatial frequency of at least 25 cycles per degree.
5. The method of claim 1 in which the segmented arc region has a
lower spatial frequency of at least 30 cycles per degree.
6. The method of claim 1 wherein the two-dimensional sub-region
comprises an area less than 20% of the color imagery.
7. The method of claim 1 wherein the two-dimensional sub-region
comprises an area less than 10% of the color imagery.
8. The method of claim 1 wherein the two-dimensional sub-region
comprises an area less than 5% of the color imagery.
9. The method of claim 1 wherein the two-dimensional sub-region
comprises an area less than 3% of the color imagery.
10. The method of claim 1 wherein the two-dimensional sub-region
comprises an area less than 1.5% of the color imagery.
11. The method of claim 1 wherein the auxiliary information
comprises digital watermark information steganographically encoded
in the imagery.
12. The method of claim 1 that further includes sensing said image
sub-region by said mobile device, and identifying a response
associated with said sub-region through use of the auxiliary
information.
13. The method of claim 1 wherein said altering further comprises
subtracting image signal energy in a second chrominance
channel.
14. The method of claim 13 wherein the added and subtracted image
signal energies fall predominantly within the segmented arc
region.
15. The method of claim 13 wherein the added and subtracted image
signal energies fall essentially within the segmented arc
region.
16. The method of claim 13 wherein the added and subtracted image
signal energies fall almost exclusively within the segmented arc
region.
17. The method of claim 13 wherein the added and subtracted image
signal energies fall exclusively within the segmented arc
region.
18. A method employing a portable user device having a processor
configured to perform at least certain acts of the method, and
having a camera portion for capturing imagery, the method
comprising the acts: transforming imagery captured by the camera
portion to yield corresponding data in a transform domain; applying
a template to detect therein a key in a segmented arc region of the
transform domain; receiving plural-bit auxiliary data; through use
of the plural-bit auxiliary data, determine a response
corresponding to the detected key; and initiating said
response.
19. A non-transitory computer readable storage medium containing
software instructions that, when executed by a processor of a
camera-equipped device, cause the device to perform acts including:
transforming imagery captured by the camera to yield corresponding
data in a transform domain; applying a template to detect therein a
key in a segmented arc region of the transform domain; receiving
plural-bit auxiliary data; through use of the plural-bit auxiliary
data, determine a response corresponding to the detected key; and
initiating said response.
Description
RELATED APPLICATION DATA
The present technology includes improvements to, and in different
embodiments makes use of, assignee's earlier work detailed in U.S.
Pat. Nos. 6,122,403, 6,408,082 and 6,590,996, published
applications 20060115110, 20070189533, 20080112596, 20080300011,
20090116683, 20100046842 and 20100048242, and application Ser. Nos.
12/337,029, filed Dec. 17, 2008; 12/482,372, filed Jun. 10, 2009;
12/498,709, filed Jul. 7, 2009; 12/640,386, filed Dec. 17, 2009;
12/643,386, filed Dec. 21, 2009; 12/712,176, filed Feb. 24, 2010;
and 12/716,908, filed Mar. 3, 2010. To provide a comprehensive yet
concise disclosure, these documents are incorporated herein by
reference, in their entireties.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 shows an image including plural hotspot regions encoded with
various keys. These may be regarded as YB-Dome spatial chromakeys.
Depicted are spatially overlapping "analog" chroma-encoded regions.
These are "invisible" to the human eye, but seen by a digital
watermark detector.
FIG. 1A shows another image similar to FIG. 1, with which a smart
phone can interact to yield different musical sounds.
FIG. 2 is a flow chart of a method that can be used in conjunction
with the image of FIG. 1A.
FIG. 3 is an illustration showing some of the different behaviors
that can be triggered in a smart phone by detection of one or more
hotspot regions in the image of FIG. 1. Depicted are both an
audio-mode, and a HotMap (GLO) mode.
FIG. 4 shows that user interaction with imagery encoded in
accordance with the present technology can employ motion.
FIG. 5 illustrates decomposition of an image into luminance and
dual chrominance channels in the spatial frequency domain, and
embedding of a key in the Y-B channel. Particularly detailed is
allocation of the YB-Dome signal space, including how such signal
can take over certain frequency bands--with no need to share with
existing artwork.
FIG. 6 is a chart showing the sensitivity of the human visual
system to the three channels of FIG. 5, as a function of image
spatial frequency.
FIG. 7 illustrates that plural different keys can be encoded within
a single chrominance channel.
FIG. 8 shows that a sub-region of smart phone's field of view can
be used to detect keys in imagery.
FIGS. 9A and 9B show how a key may be warped by perspective or
other distortion of the image, allowing such distortion to be
determined.
FIGS. 10A and 10B show different spectral signal energies within a
segmented arc in Fourier space that can be employed to represent
keys.
FIGS. 11A-11C illustrate that keys can have harmonic relationships,
and can be used in connection with binary encoding.
FIG. 12 shows another key representation in Fourier space.
DETAILED DESCRIPTION
Referring to FIG. 1, an image 10 is encoded to define one or more
regions (CK1, CK2, etc.) that can be sensed by a suitably-equipped
mobile device (e.g., a smart phone), but are imperceptible to
humans. When such a mobile device senses one of these regions, it
takes an action in response.
The content of image 10 can be arbitrary. Details (except the
hidden regions) are not shown for clarity of illustration.
The regions can be of any shape; only a few are illustrated. As
indicated, they may overlap in layered fashion. Each region has a
key associated with it (e.g., CK1, CK2, etc.). Because the
preferred embodiment employs a chrominance-based embedding
arrangement, these keys may be termed chroma keys. However, other
embedding arrangements can alternatively be used.
FIG. 1A shows an exemplary illustration, employing five regions,
hidden in the iconic Iwo Jima photograph but shown as visible for
explanatory purposes. When the left-most region is sensed by a
smart phone, it renders a musical note at the pitch "C." The
adjacent region causes the phone to render a pitch at "E."
Similarly with the next one, at "G."
The top-most region similarly causes the phone to render a musical
note at the pitch "C." Beneath it, however, is a region that causes
the phone to render the musical note E.sup. . Below is the "G"
region noted above.
If a user sweeps a suitably-programmed smart phone horizontally
across the image from left to right, so that the phone's camera
"sees" the horizontal row of regions in sequence, a C-major chord
progression is sounded. Conversely, if the user sweeps the phone
vertically, downwardly, across the indicated regions in the image,
a C-minor chord progression is sounded.
More particularly, in this example the regions are encoded with
four different keys. The same key (CK1) is used for both the
left-most and top-most regions, since they both correspond to the
same musical tone. Different keys CK2, CK3 and CK4, are used for
the other three regions.
In the illustrated example, the image also is digitally watermarked
(DWM) in known fashion to steganographically convey a plural-bit
payload of auxiliary data (e.g., 32- or 128-bits)--although this is
not always required. These plural bits can identify a data
structure that is consulted by the smart phone in determining what
response should be associated with which key.
Identification of the data structure by the plural bit auxiliary
data can be by various means. One is a pointer (e.g., an address)
to a local or remote table, or database record. In this instance,
the data structure indicates that if CK1 is detected, the phone
should render a musical tone of "C" (i.e., 262 Hz). A complete
table may instruct the phone to respond to different keys as
follows:
TABLE-US-00001 TABLE I CK1 Render 262 Hz tone CK2 Render 330 Hz
tone CK3 Render 392 Hz tone CK4 Render 311 Hz tone
This table is exemplary only, and the illustrative responses are
more basic than may often occur--simply for expository
convenience.
A tone may sound for a fixed duration, e.g., 1 second, commencing
with initial detection of the corresponding key region, or it may
persist for as long as the region is within the field of view of
the smart phone camera. In the latter case, chords of plural notes
can be rendered by positioning the camera so that it can view
some--or all--of the depicted regions. The chord is augmented with
more notes as the camera is moved away from the image, and resolves
to fewer notes as the camera is moved closer.
The use of auxiliary data to indicate what responses should be
associated with different keys allows simple dynamic reprogramming
of the user experience, even with a "fixed" image. For example, by
simply changing data in the table to which the illustrative
auxiliary data (watermark) points, the image can trigger a wholly
different set of responses--all without modification of the
imagery.
A flow chart detailing the just-described method is shown in FIG.
2.
FIGS. 3 and 4 further illustrate the just-discussed feature, by
which a smart phone camera detects different hidden regions in an
image, and can yield different outputs as the phone is moved.
"GLO" in FIG. 3 is an acronym for Graphic Latent Overlay. This
technology is detailed in documents including published patent
specifications 20080300011 and 20090116683, and refers to
arrangements in which detection of a first feature (commonly a
steganographically encoded feature) in mobile phone-captured
imagery triggers display of a second graphical feature on the phone
screen--overlaid on the image in which the first feature was
detected. In some instances, the overlaid feature is presented on
the screen at a position that is dependent on the position of the
first feature in the phone's field of view. The overlaid feature
may also be warped to correspond with apparent warping of the first
feature.
Related technologies are detailed in application Ser. Nos.
12/640,386 and 12/712,176, where such overlaid features are termed
"baubles" (and may be based on visible and/or steganographic
features of captured imagery).
Thus, instead of playing a musical tone when a key region is
sensed, the phone can respond by overlaying a graphical display on
the captured imagery.
Of course, playing musical tones and overlaying graphical displays
are a few of countless behaviors that detection of a key region can
trigger. Any imaginable action or script can be similarly triggered
when one or more of the hidden regions (which may be regarded as
"hotspots") is sensed.
Consider, as a particular example, a photo of members of a hockey
team. The oval region of each player's face is marked with a
different hidden key, CK1-CK20. When a user directs the smart phone
camera to a particular player's face, the corresponding key is
detected. A watermark is also detected from the image (and may span
the entire image). The watermark payload points to a table, like
Table I above, having an entry for each of keys CK1-CK20. This
table, in turn, may store links to Flash video presentations,
resident at an NHL web server or in the cloud, that show season
highlights for that respective player. When the user moves the
smart phone camera to view different faces in the team picture,
different video plays from the past season are rendered on the
user's phone.
(The watermark payload can serve other purposes as well. For
example, the payload can include one or more flag bits instructing
the smart phone not to provide a response if plural keys are
detected within the captured imagery. Alternatively, they may
instruct the smart phone to prompt the user to more closely direct
the camera to one particular feature (face) in the image. All prior
art uses of watermark signaling can also be employed in conjunction
with the present arrangements.)
While images are commonly perceived in the spatial (or pixel)
domain, a number of other representations are also useful, as is
familiar to those skilled in the art. Generally, these other
representations are termed "transform domains." One common
transform domain is Fourier space (commonly called the spatial
frequency domain, when used with images). By applying a Fourier
transform to an image (either as a whole, or more commonly as
divided into square regions, e.g., 8.times.8 pixels or 32.times.32
pixels), a representation of the image in Fourier space can be
obtained. Such representation is commonly presented on a complex
half-plane coordinate system, as shown in FIG. 5.
The left part of FIG. 5 shows a representative spatial frequency
domain representation of a sample image (e.g., the Iwo Jima image
earlier presented). This image is encoded with a prior art digital
watermark that conveys auxiliary data. (A suitable watermark is
detailed in U.S. Pat. No. 6,590,996.) The watermark signal energy
is interspersed among the rest of the image energy, and is not
separately identifiable in FIG. 5.
The image may be divided into component image planes, such as
luminance, and dual chrominance channels: R-G (red-green) and Y-B
(yellow-blue)--each of which has its own Fourier space
representation. These three channels, in Fourier-space, are shown
at the right side of FIG. 5.
In the illustrated embodiment, the encoding of the keys into
regions of the image is achieved by adding energy in one of the
chrominance channels (e.g., Y-B). This can be done without
objectionable visible effect because the human visual system is
less sensitive to color than it is to luminance.
This is shown in FIG. 6, which plots HVS sensitivity for luminance,
red-green chrominance, and yellow-blue chrominance, as a function
of spatial frequency, in cycles per degree. (This chart is based on
data collected at a viewing distance of 12 inches.) As can be seen,
HVS sensitivity drops off sharply at chrominance spatial
frequencies above 10 cycles per degree (cpd), and is down
dramatically at 20 cycles per degree. Sensitivity above 25 or 30
cycles per degree is essentially nil.
In contrast, image sensors used in cameras can commonly distinguish
chrominance features at such spatial frequencies.
Thus, to encode a key in an image region, such as CK1 in FIG. 1 or
1A, energy can be added to the yellow-blue chrominance channel,
desirably at frequencies of above 10, 20, 25 or 30 cpd--where the
effect is generally not noticeable to human observers, but can
readily detected from optical sensor data.
In the preferred embodiment the added energy needn't be exclusively
above the indicated frequency. But desirably most of the added
energy is in that range (e.g., more than 50%, 75%, 95%, or
98%).
In this embodiment, and in others, the magnitude of the added
energy can also be scaled to achieve a desired level of visibility
(or invisibility). In other embodiments, signal energy at
frequencies primarily below 10 cpd can be employed, if the
magnitude of the added energy is low enough to be visually
un-objectionable.
One example of suitable encoding, in Fourier space, is shown in the
lower right of FIG. 5 by the black bands. This added signal energy
dominates existing image energy in these spectral frequencies
(including phase), but the effect typically is essentially
imperceptible to human viewers. The effect is to slice out a region
in chrominance space and allocate it to representing a hidden a
key. Thus encoded, a suitably-equipped smart phone detector can
quickly discover the presence of the hidden region in the image,
and determine its extent, and respond to such discovery.
In many instances it is desirable to be able to distinguish
different keys, such as CK1, CK2, etc., in FIG. 1. This can be done
by representing different keys with different Fourier domain zones
of signal energy in the chosen (e.g., Y-B) chrominance channel
(including phase). Such arrangement is shown in the upper right of
FIG. 7.
In particular, this spatial frequency domain plot shows 10 keys,
150a through 150j. In FIG. 1, region CK1 can be encoded by adding
signal energy corresponding to the keystone-shaped Fourier zone
150a to that image region. Likewise, image regions CK2-CK5 can be
encoded by adding energy at spatial frequencies indicated by
keystone zones 150b-150e. Five additional keys can be represented
by adding energy corresponding to the spatial frequency zones
labeled 150f-150j. (It will be recognized that the discussed plot
encompasses ten different keys. A particular region of the imagery
would not necessarily have this depicted signal composition in
Fourier space. More likely, only one or a few of the keystone
features would be present.)
Again, the illustrations do not show phase, which is a helpful
dimension of these hidden signals. The phase component contributes
certainty to detection of these keys (i.e., increasing
signal-to-noise ratio).
In one particular embodiment, different keys may be of opposite
phase polarity (analogous to +1/-1). One key, CK-a, may be
represented by the keystone zone 150a with a certain phase profile
across the zone, and another key, CK-b, may be represented by
energy in the same keystone zone 150a, but with a phase profile
that is 180 degrees out of phase relative to key CK-a. This results
in a doubling of the keyspace (e.g., the ten zones 150a-150j can
represent 20 distinguishable keys).
A further multiple-key arrangement is shown in the lower right of
FIG. 7. Here N mutually-orthogonal signals in Fourier space are
employed. For purposes of illustration, a simple example is for one
signal to be a sine wave at an image frequency f. Another may
comprise a cosine signal at the same frequency. A third may
comprise a sine signal at a frequency 2f. Etc. (The design of
orthogonal signals in Fourier space is within the capability of the
artisan, although such signals are difficult to represent in simple
drawings. In one view, two keys are orthogonal if the product of
their phase profiles yields zero.)
Typically, all keys in an image are encoded at the same magnitude,
usually at a level that renders the keys imperceptible to humans in
the rendered imagery. Having a uniform magnitude across all keys
can be useful in detection, since detection of one key provides
magnitude information that can be used to help discriminate the
presence of other keys.
In some embodiments, certain keys may have different magnitudes
(e.g., they may be partially "on"). This can be used to further
expand the encoding space. E.g., key CK-c can have a magnitude that
is half that of key CK-d, but the keys are otherwise the same in
spectral frequency content and phase profile. Instead of binary
values of single keys (e.g., +phase and -phase), quaternary and
other multi-valued symbols can be used.
If the relations between different magnitudes are known by the
detector (e.g., some keys are 100% "on"; other keys are 50% "on,"
and still others are 25% "on,"), this knowledge can again be used
as a basis in discriminating keys. (These may be regarded as
"analog keys.")
In a system employing multiple keys, each can trigger a different
action (e.g., different tones in the FIG. 1A example). Still
further actions can be triggered when multiple keys are overlaid in
an image region (e.g., where CK5 overlaps CK1 in FIG. 1). Thus, in
a 10 key system, there may be ten responses corresponding to the
ten individual keys. Nine further responses can correspond to
regions where key 1 is overlaid with one of keys 2-10. Eight
further responses can correspond to where key 2 is overlaid with
keys 3-10. Seven further responses can correspond to where key 3
coexists with keys 4-10. Etc. Thus, in a system that contemplates
only layers of up to two keys, 55 different states can be defined.
(A geometric increase of possible states naturally occurs if three-
or more keys are overlaid.)
If the keys have different magnitudes, the responses triggered by
the keys can vary accordingly. In the tonal example of FIG. 1A, for
example, if key CK3 is half the magnitude of the other keys, then
its corresponding tone may be rendered at half the volume of those
triggered by the other keys.
The smart phone detector can apply Fourier-domain templates for the
different keys it expects to encounter--looking for the existence
of the hidden signals. This can comprise a correlation detector
that performs multiplications with each of the respective key
templates--judging whether resultant output signals exceed a
threshold value.
The template(s) can encompass more of Fourier space than the
precise boundaries of the sought-for key(s), allowing detection of
keys in the case of distortion of the imagery (with consequent
distortion of the keys in Fourier space).
Typically, smart phone cameras have a rectangular field of view
(e.g., 2048.times.1536 pixels). In some instances it is desirable
for the smart phone camera to be responsive to keys only in a
sub-region of this field of view. Such an arrangement is shown in
FIG. 8, in which the full camera field of view is the image frame
190, yet responses are only triggered by keys found within the
sub-region 192. The shape of the sub-region can be arbitrary; the
illustrated rectangle (of about 250 pixels in width by about 900 in
height) is only exemplary.
In another arrangement, the sub-region 192 is circular or oval.
Moreover, the software that controls use of sampled image data from
region 192 can apply a foveal filter: using all the pixels at the
center of the region for most acute detection, but using a sparser
sampling of pixels near a periphery of the region 192, so that
sensitivity or acuity tapers off.
A camera-equipped detector using the present technology can discern
the spatial extent of key regions, e.g., identifying region
boundaries to an accuracy of tens of imaged pixels or less (e.g., 5
to 10 "waxels"--watermark signal elements). The detector can thus
"image" the hidden keys, e.g., generating a virtual map of their
locations within the imagery.
If the geometrical shapes of the keys are known in advance (e.g.,
if keys are circular, or square, or equilateral triangles), then
their shapes in the captured imagery can be used to characterize
any off-axis viewing of the image. This information, in turn, can
be used to aid detection of any watermark signal encoded in the
imagery.
For example, the device processor can re-sample the image data in
correspondence with distortion of the known key shapes. If the key
is known to include two parallel edges of the same length, yet a
first edge appears in the captured imagery to be 50% shorter than a
second edge, then image points along the first edge can be
resampled by the processor at a spatial frequency twice those alone
the second edge--with interpolated sampling rates being used along
intermediate parallel lines. The watermark decoding process can
then be applied to the resampled image--from which this aspect of a
perspective warp has been effectively removed.
A more sophisticated approach to dealing with perspective does not
require knowledge of geometrical key shapes. Rather, it relies on
knowledge of a key's Fourier domain representation.
To illustrate, an original image may include a key having a Fourier
domain shape shown in FIG. 9A. If the image is captured from an
off-axis perspective, the key shape may bulge, e.g., resulting in
the Fourier domain representation shown in FIG. 9B. The
transformation, from FIG. 9A to 9B, reveals the apparent
transformation of the image as viewed by the camera (e.g., affine
and perspective warping).
A programmed processor (e.g., in a smart phone) can perform a
brute-force search for an arc-like feature in Fourier space, and
then sleuth corresponding warp parameters of the image from the
shape of the discovered arc. Another approach employs
Fourier-Mellin-based processing to locate the distorted arc zone
within Fourier domain image data, using technology akin to that
that detailed in U.S. Pat. Nos. 6,408,082 and 6,590,996.
The detailed keys are typically conspicuous in Fourier space,
allowing ready determination of warp parameters by such methods--at
least in a gross sense. Once a rough sense of the image viewing
perspective is thereby discerned (e.g., to within 5 or 10 degrees),
the image data can be resampled accordingly, e.g., as if captured
from an on-axis camera perspective. Then calibration signals
commonly found in digital watermark signals (e.g., per the
just-cited patents) can be used to resolve remaining small
distortions in the resampled imagery, allowing accurate reading of
large payload watermark data.
The artisan will recognize that calibration signals in the cited
watermark technology serve as a search-space accelerator. The key
technology detailed herein can be viewed as a supplemental
calibration signal, and detectors may be tuned to search off-axis
camera angles. Keys for this purpose can take the form of paired
staggered strings of pearl-like-features in the Fourier Y-B
magnitude space (i.e., two concentric rings of peaks in the Fourier
domain, such as shown in FIG. 12). The detector can then apply a
family of templates--corresponding to these key features (or other
known key features) as they would appear in the Fourier domain if
the image is viewed in stepped angles off-axis (e.g., in 5 degree
steps, from 0 to 45 degrees) and employing Fourier-Mellin
analysis.
This operation is suitable for quick performance by a cloud-based
processor, e.g., in 250 ms or less. The smart phone can send a
32.times.32 pixel region from the center of the camera view to the
cloud processor (either in the pixel domain, or after transforming
to Fourier space--possibly with compression that does not corrupt
the watermark signal). The cloud processor, in turn, applies the
cited templates to the received data to locate the key signal, and
returns an estimate of the viewing angle to the phone. The phone
then resamples the image in accordance with this estimated viewing
angle, and applies the cited watermark decoding process to extract
the watermark payload.
(Other divisions of labor between the phone and cloud are likewise
possible, e.g., with the cloud doing all of the processing
including extracting the watermark payload.)
In some embodiments, the keys can be designed to have "fragile"
properties, so as to aid in image/document authentication. For
example, the hidden keys can be composed of spectral frequencies
that are at or near the high frequency limit for imaging sensors to
detect. The magnitudes of these signals can also be tailored so
that the keys are barely detectable in the original image. If such
an original image is scanned and re-printed, this low amplitude
and/or high frequency data will likely degrade so that it is no
longer detectable. CCD, CMOS and other image sensors can be
designed with such technology in mind, to achieve desired
properties in the combined system.
One application of this technology is as a keypad, e.g., akin to
the sorts conventionally used to open a safe, set alarms, etc. For
example, a smart phone can capture a frame of imagery from print
media or another screen and, by reference to a watermark decoded
from the imagery or other auxiliary data, overlay a graphic showing
a touchpad. The user can touch the different buttons displayed on
the screen. The touch-sensitive screen provides output coordinate
data by which pixel regions in the captured imagery corresponding
to these touches can be determined. Hidden keys detected at these
regions in the captured imagery are noted, and the sequence of such
keys defines the "combination" or other code entered by the user
(e.g., CK4, CK2, CK5, CK9). If the entered code matches a stored
sequence, an action can be taken. Or other action can be taken in
accordance with the keystrokes (e.g., reprogramming a thermostat,
such as detailed in application Ser. No. 12/498,709).
In a variant embodiment, the outlines of the hidden keys are
discerned from the captured image, and made visible as graphic
overlays. Again, these can be touched in a desired sequence. By
reference to the keys, in conjunction with the watermark,
responsive action can be taken.
In a further variant, the graphic overlay can change as further
keys are touched.
CONCLUDING COMMENTS
While this specification earlier noted its relation to the
assignee's previous patent filings, it bears repeating. These
disclosure materials should be read in concert and construed as a
whole, together. Applicant intends, and hereby expressly teaches,
that features in each disclosure be combined with features in the
others. Thus, for example, the arrangements and details described
in this specification can be used in variant implementations of the
systems and methods described in application Ser. Nos. 12/640,386,
12/712,176, and 12/716,908, while the arrangements and details of
those patent applications can be used in implementations of the
systems and methods described in the present specification.
Similarly for the other noted documents. Accordingly, it should be
understood that the methods, elements and concepts disclosed in the
present application be combined with the methods, elements and
concepts detailed in those cited documents. While some have been
particularly detailed in the present specification, many have
not--due to the large number of permutations and combinations, and
the need for conciseness. However, implementation of all such
combinations is straightforward to the artisan from these
teachings.
Having described and illustrated the principles of the technology
with reference to selected examples, it should be appreciated that
the technology is not so limited.
For example, while described in the context of static imagery, the
technology is likewise relevant for use with video, e.g.,
comprising many successive frames, or fields.
Static images can take any known form, e.g., printed, or presented
on an electronic display screen.
Although the detailed arrangement employs digital watermarking to
convey auxiliary information by which actions corresponding to
sensed key regions can be determined, this is not necessary. Other
communications means can be utilized, e.g., RFID chips, barcodes,
header data, etc. Alternatively, a fingerprint of the imagery can
be calculated, and matched with reference information in a database
to identify the image, which identification can also serve to
identify metadata by which associations between the keys and
appropriate responses can be determined. (Examples of image/video
fingerprinting are detailed in patent publications 7,020,304
(Digimarc), 7,486,827 (Seiko-Epson), 20070253594 (Vobile),
20080317278 (Thomson), and 20020044659 (NEC).)
In still other arrangements, the auxiliary data is not necessary.
Instead, the different keys can be ascribed meaning by a known
reference, such as a data table stored in the smart phone or
elsewhere--not pointed-to or otherwise indicated by auxiliary data.
When a key is detected, the reference is consulted to determine
what behavior should be performed.
While the detailed arrangement employed a Fourier transform (e.g.,
an FFT), any number of other image transforms may alternatively be
employed. Examples include DCT, wavelet, Haar, Hough, etc.
It should be recognized that tonal signatures, such as the chord
progression illustrated in FIG. 1A, can be used as a quick check of
the authenticity, or other attribute, of an image.
Although reference was made to the keys being hidden or invisible,
this is all a matter of degree, which can be tuned to meet the
needs of particular applications. In some applications, for
example, some visibility is acceptable.
As used herein, "primarily" means more than half (50%).
"Predominantly" means more than 75%. "Essentially" means more than
95%. "Almost exclusively" means more than 98%, and "exclusively"
means 100%.
A segmented arc region refers to areas like that shown in solid
black FIGS. 5 and 6. These zones do not encompass the origin of the
complex plane--so if mirrored about the axes, there would be a gap
at the center. Put another way, a segmented arc region typically is
defined, in part, by an arc segment that serves as part of a
bounding perimeter, a non-zero distance away from the
origin--corresponding to a minimum image frequency. Practically
speaking, the segmented arc region is likewise bounded by an outer
perimeter that is a finite radius from the origin--indicating the
maximum image frequency. Likewise, a segmented arc region does not
encompass any part of the horizontal (u) or vertical (v) axes. In a
particular implementation there may be one segmented arc region, or
there may be several. (In many instances such regions may be
mirrored around an axis, such as the v axis.
The added/subtracted signal energy is typically placed within a
segmented arc region. However, it need not fully occupy the region.
An example is shown in the transform domain plane 100 of FIG. 9A.
The arc region 102 is defined by an inner perimeter 104 and an
outer perimeter 106. This region 102 does not include the origin
108. Nor does it include the horizontal or vertical axes 110, 112.
Within this segmented arc region 102 are one or more spatial
frequency zones 114 that are increased in spectral energy. Each
such zone typically comprises more than a single frequency (e.g.,
an impulse function), although a single frequency can be used in
variant embodiments. Each zone commonly has extent in at least one,
and usually two, directions in the transform plane.
Although the transform domain signal energy zones shown in FIG. 9A
have certain symmetry (e.g., around axis 116), this is not
required. FIG. 9B, for example, shows another arrangement.
Non-symmetrical arrangements are desirable in various
instances.
In some instances, the various spatial frequency zones used in
encoding an image can have harmonic relations. FIGS. 10A and 10B
show one such arrangement. FIG. 10A shows a first zone 130. FIG.
10B shows a second zone 132. Zone 132 comprises frequencies that
are twice the frequencies of zone 130. Further harmonics may
similarly be used.
Likewise, the use of different zones can be used for binary signal
encoding. The zone 130 may represent a least significant bit, e.g.,
corresponding to "1" in decimal. Zone 132 may represent a
next-least significant bit, e.g., corresponding to "2" in decimal.
The addition of signal energy in both zones, as in FIG. 10C thus
corresponds to "3" in decimal, etc.
In a variant embodiment, signal energy added to one chrominance
channel is matched, in a complementary fashion, by subtraction of
signal energy from the other chrominance channel. Such form of
embedding is detailed in patent application Ser. No. 12/337,029.
This approach allows use of lower frequency chrominance signals
without visible degradation, and yields increased signal-to-noise
ratio in the detector when processed according to the cited patent
application's teachings. (In video according to this technique, the
complementary subtraction operation can be applied to the same
frame, or to a next frame in the video sequence.)
While it is possible to mark an entire image with a single key,
more typically a sub-part of the image is so-marked. Commonly less
than 20% of the image area is marked with any particular key. More
typically (e.g., as shown in FIGS. 1 and 1A), a contiguous key
region encompasses less than 10%, 5%, 3% or 1.5% of the entire
image frame.
In practical application, artwork authoring tools such as Adobe
Illustrator and the like will doubtless be employed to add keys to
imagery. The hidden keys can be represented in visible fashion on
the user's screen, such as by dashed outlines, distinctive
cross-hatching, etc., so as to indicate the shapes to the user.
These can be dragged and resized in known fashion, using tools such
as have been popularized by Adobe. The authoring software can
further include a print-preview feature that renders the imagery on
the screen, including the keys in their hidden form--approximating
any color space transformation and other distortions inherent in
the printing workflow. (Different print-preview options may be
provided, dependent on the type or fidelity of the print process,
e.g., high resolution/fully saturated; low resolution/low
saturation, etc.) The software may include a viewing mode in which
two views (authoring and print-preview) are presented
side-by-side.
Corresponding test application software can be provided for a smart
phone used by the artist, which senses the added keys from imagery,
and provides quantitative data back to the artist. As changes are
made to the artwork, this software application can provide
immediate feedback. Diagnostic windows can be displayed on the
smart phone screen, or the authoring terminal (to which the phone
may be linked wirelessly, or by USB or the like). These diagnostic
windows can indicate key signal strengths, apparent sizes of key
regions, confidence levels for key detection, etc.
Repeated reference was made to a smart phone, yet it will be
understood that any device can serve as a detector of such signals.
The earlier-cited documents detail some of the variety of devices,
and diverse hardware configurations, that can be employed in such
systems.
To provide a comprehensive yet concise disclosure, the documents
referenced herein are incorporated-by-reference in their
entireties, as if appended hereto as supplemental parts of this
specification.
In view of the many embodiments to which principles of this
technology can be applied, it should be recognized that the
detailed embodiments are illustrative only and should not be taken
as limiting the scope of my inventive work. Rather, I claim all
such embodiments as fall within the scope and spirit of the
following claims, and equivalents thereto. (These claims encompass
only a subset of what I regard as inventive in this disclosure. No
surrender of unclaimed subject matter is intended, as I reserve the
right to submit additional claims in the future.)
* * * * *